The SciPy Stack. Jay Summet

Size: px

Start display at page:

Download "The SciPy Stack. Jay Summet"

Patricia Gordon
6 years ago
Views:

1 The SciPy Stack Jay Summet May 1, 2014

2 Outline Numpy - Arrays, Linear Algebra, Vector Ops MatPlotLib - Data Plotting SciPy - Optimization, Scientific functions TITLE OF PRESENTATION 2

3 What is Numpy? 3rd party module that extends the Python language Matrix and linear algebra operations General support for N-Dimensional Arrays Heterogeneous elements Direct mapping to C/C++/Fortran memory buffers Matrix operations Vectorized Algorithms Bindings to C and Fortran matrix math libraries Requires compiled support libraries Installers are platform specific 3

4 Importing Numpy Must install it via the SciPy stack or individually Most people import the numpy library as np Arrays can be created from lists or nested lists >>> import numpy as np >>> x = np.array( [1,5,10] ) >>> x array( [1, 5, 10] ) >>> y = np.array( [ [1,2,3], [4,5,6] ] ) >>> y array( [ [1,2,3], [4,5,6] ) 4

5 Data Types Numpy stores all data in an array using a single data type Direct mapping to memory / space required The Python data types have default mappings You will most commonly use: bool - boolean, stored as a byte int_ (Same as C long, either int32 or int64, architecture dependent) float_ ( float64) complex_ (complex128) Will convert python int/float/bool/complex automatically 5

6 Data Types Numpy supports many different native data types Will convert python int/float/bool/complex automatically The.dtype attribute of an array tells the actual data type Can convert to another data type using.astype() >>> x = np.array( [1,5,10] ) >>> x array( [1, 5, 10] ) >>> x.dtype dtype( int32 ) >>> y = x.astype(np.float64) >>> y array( [ 1., 5., 10.] ) >>> y.dtype dtype('float64') 6

7 Numpy Array Creation Convert python array-like data structures Using built in array creation methods: np.zeros ( shape ) np.ones( shape ) np.arange(start, Stop, step) - Like python range np.linspace( Start, Stop, number) Most accept an optional dtype parameter >>> x = np.zeros( (2,2), dtype=np.int_ ) >>> x array( [0, 0], [0, 0] ) >>> y = np.linspace( 1, 4, 6 ) >>> y array( [ 1., 1.6, 2.2, 2.8, 3.4, 4. ] ) 7

8 Array = Block of Memory All data in the array is stored in a single block of memory The.shape attribute stores the size of each dimension You can change the shape on the fly! must retain the exact number of elements A 1 x 10 array and a 2 x 5 array both have 10 data items >>> x = np.arange(10) >>> x array( [0, 1, 2, 3, 4, 5, 6, 7, 8, 9] ) >>> x.shape (10, ) >>> x.shape = (2,5) >>> x array( [ 0, 1, 2, 3, 4], [ 5, 6, 7, 8, 9] ] ) 8

9 Array Indexing Works like indexing in python Zero based negative indexes count from the back BUT, You CAN index all dimensions at once if you want! >>> x = np.arange(10) >>> x[-1] 9 >>> x.shape = (2,5) >>> x[1][3] 8 >>>x[1,3] 8 9

10 Array Slicing Works like indexing in python Provide a Start:Stop:Stride Can do this for multiple dimensions at once >>> x = np.arange(25).reshape(5,5) >>> x array([[ 0, 1, 2, 3, 4], [ 5, 6, 7, 8, 9], [10, 11, 12, 13, 14], [15, 16, 17, 18, 19], [20, 21, 22, 23, 24] ] ) >>> x[0:5:2, 2:4] array([[ 2, 3], [12, 13], [22, 23] ] ) 10

11 Array Slicing - Aliasing! Slices provide VIEWS Like an alias of a list See array indexing if you want to copy a subset of data! >>> y = x[0:5:2, 2:4] >>> y[0,0] = 999 >>> y array([[999, 3], [ 12, 13], [ 22, 23]]) >>> x array([[ 0, 1, 999, 3, 4], [ 5, 6, 7, 8, 9], [ 10, 11, 12, 13, 14], [ 15, 16, 17, 18, 19], [ 20, 21, 22, 23, 24]]) 11

12 Advanced Indexing You can specify a set of indices with an array to select a copy of the data. >>> x array([ 0, 5, 10, 15, 20, 25]) >>> y = x[ np.array( [1,3,5] ) ] >>> y array([ 5, 15, 25]) >>> y[0] = 99 >>> y array( [99,15,25] >>> x array( [ 0, 5, 10, 15, 20, 25] ) 12

13 Advanced Indexing BUT, if you assign to an array indexing operation, it modifies the original... >>> x array([ 0, 5, 10, 15, 20, 25]) >>> y = x[ np.array( [1,3,5] ) ] >>> y array([ 5, 15, 25]) >>> x[ np.array( [1,3,5] ) ] += 1 >>> x array([ 0, 6, 10, 16, 20, 26]) >>> 13

14 Array Broadcasting Element-by-element operations on two arrays typically require the arrays to have the same shape A scaler is treated as a 1x1 array that gets broadcast over all elements >>> y = np.arange(1,5) >>> y array([1, 2, 3, 4]) >>> y * 2 array([2, 4, 6, 8], dtype=int32) >>> y * 2. array([ 2., 4., 6., 8.]) 14

15 Array Broadcasting Arrays that are short in a dimension can also be broadcast! The Rule: In order to broadcast, the size of the trailing axes for both arrays in an operation must either be the same size or one of them must be one. >>> a = np.array( [ [0, 1, 2], [3, 4, 5] ] ) >>> b = np.array( [4,5,6] ) >>> c = a * b #Multiplication, not np.dot(a,b) >>> c array( [ [ 0, 5, 12], [12, 20, 30] ], dtype=int32) 15

16 Vector Algorithms: Calculating y = sin(x), for 100 elements between -4 and 4 >>> x = np.linspace(-4,4,100) >>> y = np.sin(x) 16

17 What is Matplotlib? Python library that supports plotting data Many features were copied from MATLAB Easily produce Graphs/Plots of Lines Histograms Bargraphs Scatterplots 3D surfaces 17

18 Plotting sin data Plotting 2D data (x vs y) requires two lines of code Three if you count the import... import numpy as np import matplotlib.pyplot as plt x = np.linspace(-4,4,100) y = np.sin(x) temp = plt.plot(x,y) plt.show() 18

19 Filled Plot import numpy as np import matplotlib.pyplot as plt x = np.linspace(-np.pi, np.pi,100) y = np.sin(x) temp = plt.fill(x,y, color="green") plt.grid(true) plt.show() 19

20 Other plots Pie charts import matplotlib.pyplot as plt labels = [ 'Red', 'Green', 'Blue', 'Yellow' ] colors = [ 'red', 'lightgreen', 'lightskyblue', 'gold'] values = [3,4,10,1] explode = (0.2, 0, 0, 0) plt.pie(values, explode=explode, colors=colors, labels=labels, shadow=true, autopct='%1.1f%%') plt.show() 20

21 3D Plots require 3D data Numpy meshgrid function allows you to create arrays with all possible X or Y coordinates >>> x = np.arange(0,5,1) >>> y = np.arange(0,4,1) >>> x array([0, 1, 2, 3, 4]) >>> y array([0, 1, 2, 3]) >>> NX,NY = np.meshgrid(x,y) >>> NX array([[0, 1, 2, 3, 4], [0, 1, 2, 3, 4], [0, 1, 2, 3, 4], [0, 1, 2, 3, 4]]) >>> NY array([[0, 0, 0, 0, 0], [1, 1, 1, 1, 1], [2, 2, 2, 2, 2], [3, 3, 3, 3, 3]]) 21

22 3D Plots require 3D data These arrays can be used to calculate a 3rd (Z) axis based upon a formula, such as euclidean distance from origin. Note the distance at point (4,3) is 5 (3x4x5 triangle) >>> D = np.sqrt( NX**2 + NY**2 ) >>> D array([[ 0., 1., 2., 3., 4. ], [ 1., , , , ], [ 2., , , , ], [ 3., , , , 5. ]]) >>> 22

23 3D Plots require 3D data And you could calculate the sin of the distance... >>> Z = np.sin(d) >>> Z array([[ 0., , , , ], [ , , , , ], [ , , , , ], [ , , , , ]]) 23

24 3D surface Plot >>> import matplotlib.pyplot as plt >>> from mpl_toolkits.mplot3d import Axes3D >>> fig = plt.figure() >>> ax = fig.gca(projection='3d') #Requires Axes3D import! >>> surf = ax.plot_surface(nx,ny, Z, rstride=1, cstride=1, linewidth=0 ) >>> plt.show() 24

25 Surface Plot - more points import numpy as np X = np.arange(-5, 5, 0.1) Y = np.arange(-5, 5, 0.1) NX,NY = np.meshgrid(x, Y) D = np.sqrt(nx**2 + NY**2) Z = np.sin(d) import matplotlib.pyplot as plt from mpl_toolkits.mplot3d import Axes3D fig = plt.figure() ax = fig.gca(projection='3d') surf = ax.plot_surface(nx,ny, Z, rstride=1, cstride=1, linewidth=0 ) plt.show() 25

26 Adding more bling... Add a colormap and colorbar (key) Customizing the Z axis 26

27 Colormap & Z-axis scale from matplotlib.ticker import LinearLocator, FormatStrFormatter from matplotlib import cm... ax = fig.gca(projection='3d') surf = ax.plot_surface(nx,ny, Z, rstride=1, cstride=1, linewidth=0, cmap=cm.coolwarm ) #Colorbar requires the surface to have a colormap. fig.colorbar(surf, shrink=0.5, aspect=5) #Customize Z-axis ax.set_zlim(-1.01, 1.01) ax.zaxis.set_major_locator(linearlocator(10)) ax.zaxis.set_major_formatter(formatstrformatter('%.02f')) 27

28 What is SciPy? Python library that builds upon Numpy arrays Provides higher level math support for Optimization Linear Algebra Interpolation FFT Ordinary Differential Equation Solvers 28

29 SciPy Function minimization Classic example function: Rosenbrock Source code in handout: rosenbrock_demo.py 29

30 Function Minimization Define the function to accept a sequence (of parameters) def rosenbrock(parms): x,y = parms return (1-x)** * (y-x**2)**2 Source code in handout: rosenbrock_minimum.py 30

31 Function Minimization Import and call fmin from scipy.optimize Provide a starting point, and GO! import numpy as np from scipy.optimize import fmin x0 = np.array( [4,-3]) #Try starting at (4,-3) res = fmin( rosenbrock, x0, xtol=1e-8) print res Source code in handout: rosenbrock_minimum.py 31

32 Function Minimization Optimization terminated successfully. Current function value: Iterations: 90 Function evaluations: 174 [ 1. 1.] Source code in handout: rosenbrock_minimum.py 32

33 Fin Questions? 33

Programming for Engineers in Python

Programming for Engineers in Python Autumn 2016-17 Lecture 11: NumPy & SciPy Introduction, Plotting and Data Analysis 1 Today s Plan Introduction to NumPy & SciPy Plotting Data Analysis 2 NumPy and SciPy